Pre-trained GFPGAN with `sft_half=False`

Hi, thanks for open-sourcing this nice work. I was wondering if you happen to keep the ablated model where the `sft_half` is set to False, my understanding is that it will keep the identity better and might help in my case. Thanks in advance!